Will the Private Equity AI Roll-Up Play in Services Work?
Strategy
Investors are paying software-style prices for accountancy practices on a simple theory: point AI at a people-heavy business and margins go from 15% to 40%. The businesses that have actually done it are telling their shareholders the opposite.
Investors are paying software-style prices for accountancy practices, letting agents and IT support firms on a simple theory: point AI at a people-heavy business and margins go from 15% to 40%.
The businesses that have actually done it are telling their shareholders the opposite.
In April, the chief executive of HCL Technologies warned that "AI deflation" would cut his revenue by 3–5%. His counterpart at Tata Consultancy Services — India's largest IT firm, and with some 585,000 staff one of the biggest private employers on earth — called the phenomenon "degrowth". Revenue still went backwards.
India's IT sector is the most AI-capable, most process-disciplined services industry in the world. It lost $68.6bn of market value in February alone, as investors concluded that AI cuts what a services firm can charge faster than it cuts what the firm spends.
Billions are being committed to the opposite bet. General Catalyst has $1.5bn behind the strategy; Thrive Holdings raised $2bn at a $12bn valuation this month; in the UK, letting agency consolidator Dwelly raised $170m in July with EQT Growth leading. General Catalyst reckons 60% of the UK estate agency market sits in 3,000-odd firms under £3m of revenue. If this works, it reaches British professional services inside two years.
Having run private-equity-backed integrations, I ask what has to be true for this to pay. Start with the easy part: the productivity is real. AI does make professional work faster, most of all for the least experienced staff — exactly where a services firm's cost base sits. Three harder things follow.
One: you have to keep the saving
Back to the Indians. They got the productivity but could not keep it: their competitors got the same tools in the same quarter, and clients noticed.
That is a competitive market problem rather than an Indian one, and it is worse in the sectors being rolled up. If you bill by the hour or the head, efficiency is not margin — it is a discount you are about to be asked for. Nobody has explained why a mid-sized accountancy practice with four local rivals escapes what TCS could not.
Two: it has to reach the profit line
McKinsey's State of AI survey, published last week: 88% of organisations now use AI regularly, up again on the year. But only 37% say it has made any measurable difference to profit — the same as last year — and only 6% can trace 5% or more of their profits to it. Also unchanged.
More adoption, more deployment, and the profit line has not moved.
Three: your people have to help
This is the one I would worry about most, because you are asking staff to automate their own hours.
Accounting Today's survey of firms bought by private equity found more than two thirds of partners satisfied — and 15% of everybody else. Partners received consideration; staff received a utilisation target. It is the second group the AI plan depends on, while the selling partners sit on deferred payments tied to old revenue that reward them for changing nothing for two or three years.
It can be done well: Schellman reports attrition under 4%, having protected remote working through its sale. But that is a firm being careful, not one being transformed.
And the timetable has to fit the debt
Rough arithmetic, mine rather than anyone's research: borrow four and a half times profits at today's rates of around 9%, and roughly 42p in every pound of profit leaves as interest from day one. Add integration costs, which EY puts at 3.5–10% of the acquired firm's revenue.
Against that, Everest Group's founder reckons the 40% savings sold on six-to-eighteen-month timelines are real but take three to five years. And Bain notes a deal that needed 5% annual profit growth in the 2010s needs 10–12% now, because rising valuations no longer do the work.
The savings arrive towards 2030. The interest arrives on Monday.
Then who buys it?
Accounting has consolidated since 2021 — around 24 of the top 100 US firms now carry PE money — so the first exits are arriving. They are sales to other private equity firms: Citrin Cooperman from New Mountain to Blackstone, Schellman from Lightyear to Goldman, with the broker of many of these deals expecting "a lot more in 2028".
Not trade buyers. Not public markets. The next sponsor. That works while the music plays, and tells you nothing about whether the software-style valuation at the end of the chain exists at all.
So — will it work?
In a narrow shape, probably. But note who is most sceptical.
FTI Consulting asked more than 500 private equity decision-makers to rank ten value-creation levers. M&A came last: only 9% made it a priority, and 67% said it takes over a year to deliver the value expected. AI ranked ninth of ten on how often it is actually used, despite 42% calling it the biggest force reshaping the industry. The two levers this thesis rests on are, by the industry's own account, the least effective and the least used.
The longer-run evidence agrees on shape. BCG's buy-and-build study — the most detailed we have, though its deals were exited by 2012 — found the strategy beat standalone deals, 31.6% a year against 23.1%. But programmes with more than two bolt-ons returned 19.9%, worse than doing nothing clever. Small platform, few acquisitions, operators in charge, ideally where you set the price rather than quote against three rivals. Not a thirty-deal programme.
In fairness: not one AI roll-up has failed. No write-down, no down round, nothing. They are three years old and value themselves, so we won't know until 2029. I am reasoning from structure, not wreckage, and you should weigh it accordingly.
The AI is the easy part. It always is.